VOICE ASSIMILATION IN MARATHI

This article discusses voice assimilation phenomenon in Marathi, an Indo-Aryan language. The goals of this study are three-fold: i) to provide empirical evidence for regressive assimilation in Marathi as claimed in previous studies by Junghare (1969), ii) to check whether obstruents and breathy voice can undergo and/or trigger voice assimilation, iii) to investigate what type of language Marathi belongs to based on Wetzels and Mascaró’s (2001) language typology. The results show that Marathi exhibits voice assimilation as stipulated by Junghare, and it belongs to type II languages that have voice assimilation and no word-medial and word-final devoicing.

Previous study on Marathi's phonology was carried out by Junghare (1969: 41-42) who claimed that there is a regressive voice assimilation in Marathi. The voiced obstruent will undertake devoicing when it precedes a voiceless obstruent, whereas the voiceless obstruent will become voiced when it is followed by a voiced obstruent. The followings are examples of voice assimilation provided by Junghare. We can see that voice assimilation occurs in affricate-stop cluster (1 and 2), fricative-stop sequence (3), as well as stop-stop cluster (4). 1.
Ladǝto  ladto  latto '(he) loads' Junghare (1969) also described that the retroflex stop [ḍ] does not assimilate in voicing. When the retroflex stop is found in a consonant cluster that does not agree in [voice] feature, it will change into a flap. Junghare's hypothesis regarding voice assimilation in Marathi raises some questions. From data 3, we can see that a voiced aspirate triggers the voicing of a preceding fricative. Vaux's (1998) claim, based on empirical evidence from different languages, is that voiced fricatives are specified as [-spread]. Since the specifications of a breathy voiced (voiced aspirate) are [voice] and [spread glottis], the assimilation of a voiceless fricative into a voiced fricative is nearly impossible because of the conflict in the feature [spread]. Consequently, the aims of this study can be formulated as follows: a. To investigate whether there is regressive voice assimilation in Marathi. b. To investigate which obstruents undergo voice assimilation. Can voiced aspirates (breathy voiced) trigger assimilation for the preceding fricative? c. To determine which type does this language belong to according to Wetzels and Mascaró's typology.
In section 2 of this paper, the methods of the experiment and the description of points that will be analyzed are explained. Section 3 includes the measurement of VOT average of initial stop and that it is similar to Yiddish in the typology of voicing and devoicing, as well as exemplifying the findings on voice assimilation in consonant cluster. Conclusion can be found in section 4.

II. Methods
Five native speakers of Marathi, three males and two females (ranging in age from 20 to 30), were recorded in a sound-treated room at the University of Iowa. The speakers read a list of words consisted of Marathi's four stops series (i.e. voiced, voiced aspirate, voiceless, and voiceless aspirate) available in initial, intervocalic, and final position. Consonant clusters of stops and fricative-stop were also included in the list (see Appendix). The duration of voice onset to stop release is measured for initial stops, whereas the presence of voicing for word-final stops is noted in order to check for syllable final devoicing. The analysis of the data includes: a. VOT measurement of each obstruent in initial position b. The presence of voicing in final position. c. The presence of voicing in consonant cluster: Is there regressive voice assimilation?

III. Findings and Discussion
If we look at the spectrogram of a word in Marathi below (e.g. [dǝgǝᶑ]), we can see that Marathi is a true voiced language because devoicing does not occur in word-medial or word-final position. The spectrogram below clearly shows the prevoicing of the initial stop [d], the prevoicing of the middle stop [g] and the voicing of the final stop [ᶑ] up until closure. Moreover, if we take Junghare's analysis regarding voice assimilation in Marathi into account, it means that Marathi can be categorized as a type II language, like Yiddish. (1) Spectogram of subject 1's production of the word dǝgǝᶑ 'stone' d ǝ g ǝ ᶑ Voice Onset Time was used by Lisker and Abramson (1964( , in Beckman et.al. 2011) to differentiate stops in aspirating language like German. The negative VOT indicates voiced stops, short-lag VOT signifies voiceless stops, and long-lag VOT specifies voiceless aspirated stops. The

International Journal of Linguistics and Discourse Analytics
Vol.2, No.1, September 2020

P-ISSN 2721-8899 E-ISSN 2721-8880
57 average VOT of initial stops in Marathi and Hindi was measured by Lisker and Abramson (1964( , in Berkson 2012 and the result can be seen below If we compare the results of Marathi's VOT that Lisker and Abramson calculated with the result of average VOT that I computed in table 3 below, we can see that the difference is not significant. The hierarchy of VOT from the highest to lowest is T h > T > D h > D. 77 out of 80 tokens of word initial voiced and breathy voiced shows prevoicing. The three tokens which do not exhibit voicing were three different words spoken by two different subjects (see spectrogram 2, 3, and 4 below). Therefore, this devoicing of word initial stop does not have any significance to the analysis that Marathi is a true voice language. (2) Spectogram of subject 5's production of the word bolne 'talk-3sn' This section looks at whether Junghare's claim (1969) on regressive voice assimilation in Marathi can be justified or not. Based on the examples that she provided, it seems that all obstruents (including stops, fricatives, and affricates) assimilate in voicing depending on the [voice] feature of the following obstruent, except for voiced retroflex which assimilate into a flap in a consonant cluster. I divide this section into 5, looking at different combination of consonant clusters to see whether regressive voice assimilation really occurs in Marathi.

Voiced stop-voiceless stop cluster
The following is the list of words with voiced-voiceless stop cluster. a. Nagpur = Nagpur b. puᶑke = a small packet c. toᶑte = picks.pres.3sf d. khoᶑkǝr = naughty e. chiᶑkaa = snappish f. aᶑkǝn = hitch g. waᶑhpyaa = waiter From the list, we can see that only the first word has a cluster consisted of voiced velar stop and voiceless bilabial stop. Three out of five spectrograms of the word Nagpur show that the voiced velar stop devoiced when it precedes voiceless bilabial stop. The other two showed that voicing can still be maintained in a voiced-voiceless cluster in a careful speech where the voiced stop is being released, as shown by spectrogram (5). When the first stop in the consonant cluster is not released, it is more likely that voice assimilation occurs. Notice the absence of voicing in spectrogram (6) compared to spectrogram (5). (5) Spectogram of subject 2's production of the word Nagpur (6) Spectogram of subject 3's production of the word Nagpur n a k p u r The rest of the words in the list were analyzed to prove Junghare's claim that voiced retroflex does devoiced in a consonant cluster as it will change into a flap. The result of analysis shows that this claim is true, as illustrated by spectrogram (7) and (8). Spectogram (7) portray [ᶑ] in word-initial position. If we compare it with spectrogram (8), we can see the difference that shows voiced retroflex [ᶑ] assimilated into a flap, thus devoicing is not possible.
Spectogram of subject 2's production of the word ᶑoke 'head' (8) Spectogram of subject 3's production of the word toᶑte 'pluck.pres.3sf' t o ɽ t e

Voiced affricate-voiceless stop/affricate cluster
There is only one example of voiced affricate-voiceless stop cluster in the data, i.e. kudʒka 'rotten'. Apparently, regressive voice assimilation happens in this sequence (spectrogram 10), although 3 out five speakers maintained the voicing of the first consonant in the cluster, as illustrated by spectrogram (9). In spectrogram (9), we can see the voicing along with the frication of the affricate /dʒ/, while in spectrogram (10) voicing only occurs during closure, but it does not last along with the frication. (9) Spectogram of subject 1's production of the word kudʒka 'rotten' (10) Spectogram of subject 5's production of the word kudʒka 'rotten' k u tʃ k a

Voiced affricate-voiceless affricate cluster
A sequence of voiced affricate and voiceless affricate is included in this study in order to ascertain the possibility of voice assimilation in clusters other than stops cluster. There is only one word that has this voiced-voiceless affricates sequence, which is adʒca 'of today'. Out of the five subjects, three of the outcomes show voice assimilation. It seems that Junghare's hypothesis (1969) regarding voice assimilation of voiced-voiceless affricates is correct. If you compare the spectrograms below, you can clearly see the devoicing of the first affricate in the sequence. From spectogram (11), we can notice that the voicing of the first affricate in the cluster is maintained by releasing the affricate, as indicated by the gray bar in the middle of the spectrogram. (11) Spectogram of subject 1's production of the word adʒca 'of today' The idea that voicing can be maintained by the release of the affricate or stop is also supported by the following spectrogram. When the following spectrogram is compared to the previous one, it is obvious that there is no release of affricate in between the consonant cluster, thus delinking of [voice] feature occurred as the preceding affricate agrees with the voice feature of the following consonant. (12) Spectogram of subject 3's production of the word adʒca 'of today' a t c a

Voiceless stop-voiced stop cluster
Below is the list of tokens with voiceless-voiced stops cluster. a. topᶑi = small cap of a baby b. taapdayk = annoying c. potdar = surname d. ekda = once e. ǝphgǝt = accident f. kaapᶑa = cloth g. saapdle = find (intr)-past-3sm h. tǝbǝkᶑii = flying saucer Out of 40 tokens, only 27.5% are produced with voice assimilation in the consonant cluster. Subject 3, 4, and 5 are the ones that usually assimilate the [voice] feature of the initial consonant depending on the [voice] feature of the following consonant in the consonant cluster. Voice assimilation never occurs in the speech of subject 1 and 2. From the graph below, we can see that the consonant cluster in two words, topᶑi and tǝbǝkᶑii, does not undergo voice assimilation, whereas for the other words, there are two speakers that assimilate the consonant cluster for each word in average.

Voiceless affricate-voiced stop cluster
If we refer to Vaux's claim (1998) regarding the specification of voiced fricatives which is [-spread], the fact that a breathy voiced can trigger the voicing of a voiceless fricative seems to be unfeasible as voiced fricatives and breathy voiced have opposing specification for [spread glottis]. Therefore, it is very interesting to see if voice assimilation does happen in the consonant cluster in kac+ghǝr 'glass house'. From the analysis of the spectrogram, we can see that there is no voicing of a voiceless fricative that is triggered by a voiced aspirate, as portrayed by (13)

IV. Conclusion
The empirical data support Junghare's claim that regressive voice assimilation occurs in Marathi. In addition, obstruents can trigger voice assimilation, whereas the breathy voice shows conflicting results. Moreover, from the result of the data analysis, we can conclude that Marathi is similar to Yiddish because it does not have word-final devoicing but some consonant clusters undergo regressive voice assimilation, although the data is not consistent as some speakers did not seem to have voice assimilation in pronouncing consonant cluster. Therefore, it can be assumed that Marathi conforms to the same constraints ranking as Yiddish, that is IDONSVO, AGREE >> IDVO>>*VO. However, the fact that retroflex stop is not prone to voice assimilation may or may not influence this constraints ranking. [taapdayk] तापदायक = annoying 63.